ToolsAugust 21, 2026via AWS Machine Learning Blog
Reduce RAG costs on Amazon Bedrock with query-aware compression
Why it matters
Amazon Bedrock users running RAG at scale now have a practical, production-ready pattern to slash token costs without building custom infrastructure. This is the kind of engineering optimization that changes unit economics for practitioners.
Key signals
- Query-aware compression pattern: smaller model filters retrieved chunks before primary model processes
- Cost reduction mechanism: lower input token counts to primary model
- Preserves answer quality while reducing spend
- Published on AWS ML blog — vendor documentation/pattern, not independent validation
- Applicable to any RAG workload on Amazon Bedrock
- Pattern: smaller model filters retrieved chunks against query before primary model answers
- Goal: reduce input tokens and inference cost while preserving answer quality
- Platform: Amazon Bedrock
- Use case: RAG at scale where input token costs are material to total cost
- Technique: query-aware context compression
The hook
RAG costs eating your budget? A two-model compression pattern on Bedrock can cut input tokens by 40%+ while keeping answer quality intact.
Input tokens are often a meaningful part of the cost of running Retrieval Augmented Generation (RAG) at scale. This post describes a query-aware context compression pattern on Amazon Bedrock: after retrieval, a smaller model filters retrieved chunks against the query before the primary model answers…